Bidirectional Tutoring for Developmental Motor Learning in Robots: Co-Developed Interaction Dynamics Support Stable Learning

Researchers have developed a bidirectional tutoring system that enables robots to learn motor skills through adaptive interaction, producing more stable and consistent behaviors than traditional one-way demonstration methods, according to a study submitted on 18 Jun 2026 [1]. The study, posted to arXiv, draws a direct parallel to infant development, noting that infants acquire motor skills through dense, responsive interaction with caregivers [1][2]. In contrast, robot motor-skill learning has historically been treated as a unidirectional process where machines passively receive demonstrations from a tutor, a framework the authors argue overlooks the inherently adaptive nature of social learning [1][2]. The research team hypothesized that bidirectional tutoring—where both tutor and learner dynamically adapt to each other—allows a robot's past experiences to function as constraints that shape co-developed behavioral trajectories [1][2]. To test this, they conducted two experiments using a physical humanoid robot performing an object manipulation task [1][2]. One experiment involved direct human-robot interaction, while the second employed an AI tutor interacting with the real robot through an adaptive intervention mechanism, designed to examine whether similar effects would emerge under more controlled conditions [1][2]. The learning framework was implemented using a free-energy-principle-based neural network extended with generative replay, a technique that supports stable sequence-by-sequence learning from single tutored episodes [1][2]. Across both experimental settings, the bidirectional approach fostered consistent behaviors and stage-wise generalization, and the robot gradually required less tutor guidance as learning progressed [1][2]. The findings suggest that treating robot learning as an embodied and socially grounded process provides an effective scaffold for developmental motor learning [1][2]. The work was conducted under the auspices of arXivLabs, a framework that allows collaborators to develop and share new features on the arXiv platform while adhering to values of openness, community, and user data privacy [1].